Seek for Incantations: Towards Accurate Text-to-Image Diffusion Synthesis through Prompt Engineering